Go for a walk?

Stephen J. Mraz
Senior Editor

At the Robotics Institute on the campus of Carnegie Mellon, grad student Jonathan Hurst is following the age-old engineering principle, "Keep it simple," as he explores two-legged motion systems. His initial prototype, for example, is so simple it only has a single leg and is constrained to two dimensions: up and down, and forward and back. But he is confident his prototype will let him determine the role compliance or muscle and tendon springiness (the opposite of stiffness) play in establishing walking and running gaits. Eventually, he would like to see his work applied to two-legged walking robots. But that day may still be a long way off, he says.

His prototype, dubbed BiMasc (Biped with Mechanically Adjustable Series Compliance), consists of a hip, thigh, and shin segment, three motors that wind and unwind cables, five cable differentials made of pulleys, and a pair of springs. One motor controls leg length, one controls leg angle, and the third adjusts leg stiffness, which can take place on the fly. The robot is tethered, with power and control coming through an umbilical.

The leg uses off-board computers, a Compact PCI (from One Stop Systems), and a Kontron SBC with a Pentium M 2.0-GHz processor, and 512 Mbyte of RAM. "This is probably overkill," says Hurst. "But we don't want to be limited by processor speed."

The two springs (from Gordon Composites) are made of the same composites found in compound archery bows. The springs are rectangles measuring 3 X 24 X0.25 in., and can store up to 300 J. The springs, like human muscle pairs, are set up in antagonism, pulling against one another and always in tension. This pretension, which simulates leg stiffness, is controlled by the third motor. The springs are clamped at one end and cables run from their free end around spiral pulleys. The pulleys let Hurst give the leg almost any linear or nonlinear spring function.

Many engineers, especially those with backgrounds in industrial robots, use electric motors or some other actuator that can be made to act like a spring through software. (In such a set up, the motor or actuator responds to displacement by applying a force equal to the spring constant multiplied times displacement, or f = kx, the classic spring equation.) "But these systems run into problems with impacts, as when the leg hits the ground. The motor's inertia dominates the behavior of the system and the simulated spring no longer follows the spring function programmed into the software," says Hurst.

Motors on BiMasc are custom wound (from Emotech) and were designed for high torque and small size. Horsepower is not as important as torque because when the leg is on the ground, motor speeds are low. "But they still need to generate enough torque to hold back the springs," says Hurst. "And when the leg is in the air, the leg segments must move quickly though there is little if any force on them."

The leg length and angle motors generate 30 N-m of peak torque, draw up to 30 A, and have a top speed of 1,300 rpm. The pretension motor has roughly half the torque but the same speed.

Cables and pulley-based differentials are used for their light weight, strength, zero backlash, and low cost. The differentials are limited in that they can't rotate indefinitely like gears, but that's not an issue with BiMasc. The cables (Saba Industries) are uncoated steel with high fiber counts for flexibility.

"Cables are a nontraditional approach to robot design," says Hurst. "But they let us mount the springs on the walker rather than on the leg segments where they would have to travel back and forth with the walker's movements."

Hurst's goal is to build a dynamically simple walker, one that almost walks itself on flat, level ground. He notes that some engineers have built passive dynamic walkers that have no power source, just properly distributed masses and linkages. These devices have a slow but natural walking gaits when traveling down a slight hill. "I'd like to take that idea and wrap a good control system around it," he says.

The software and controls involved in walking are relatively straightforward. But they get more complicated when handling disturbances such as bumps in the road, strong winds, or perhaps inclines. So Hurst is looking for a simple mathematical model of walking with known natural dynamics. Then controls for BiMasc can take advantage of those natural dynamics.

"I want the robot to be good at walking and running over a variety of terrains. If it looks human or animallike when it runs, it would be a fortunate coincidence, not a goal," says Hurst. This project is not about biokleptics, a term I've heard which refers to taking ideas from biology to use in robotics."

Currently, Hurst is developing a controller to handle running. Next come experiments to find the optimal stiffness or compliance for running. During these trial runs, all parameters for the gait will stay constant, except for stiffness. For example, speed, stride length, and maximum height will be held the same as Hurst changes the stiffness (through the smaller pretensioning motor). Hurst will also measure how much energy BiMasc is using. If his theory is correct, Hurst will find a specific stiffness that corresponds to the best efficiency (least energy used.).

Hurst is working on his bipedal robot with help from his adviser, Al Rizzi, as well as Professors Jessy Grizzle and Ben Morris at the University of Michigan. Funding is through the National Science Foundation.